Fast discrimination and counting of filled/unfilled rice spikelets based on bi-modal imaging

Spikelets per panicle and grains per panicle (also known as filled spikelets per panicle), directly contributing to rice yield, are two imperative traits that need to be evaluated in yield-related research. Current determination of total spikelet number and filled spikelet number are generally measured manually, which is tedious and subjective. This paper proposes a new method of counting total spikelets and filled spikelets simultaneously, based on automatic discrimination of filled and unfilled spikelets by combining visible light imaging and soft X-ray imaging. Visible light imaging was applied to measure the projected area of the spikelet hull, while soft X-ray imaging yielded the projected area of the inner brown rice kernel. The filling rate, defined as the area ratio of rice kernel to hull, was used to discriminate the filled and unfilled spikelets. 29 panicle samples were tested to evaluate the efficiency and accuracy. The results showed that the counting efficiency was approximately 2000 spikelets/min. The root mean squared error (RMSE) was 0.42 for total spikelet number and 0.77 for filled spikelet number. The mean absolute percentage errors (MAPE) were 0.22% and 0.80% for each, respectively. The method shows great potential in improving the efficiency of trait evaluation in plant breeding and genetic research, as well as serving for crop phenomics. (C) 2010 Elsevier B.V. All rights reserved.